Invent. math. 122, 531-557 (1995)
Inventiones mathematicae 9 Springer-Verlag 1995
4-Manifold topology II: Dwyer's filtration and surgery kernels Michael H. Freedman j, Peter Teichner 2 i University of California, Department of Mathematics, San Diego, La .lolla, CA 92093-0112, USA; e-mail:
[email protected] 2 Universitfit Mainz, Fachbereich Mathematik, D-55099 Mainz, Germany; e-mail:
[email protected] Oblatum 20-11-1995 & 26-V-1995
Abstract. Even when the fundamental group is intractable (i.e. not "good") many interesting 4-dimensional surgery problems have topological solutions. We unify and extend the known examples and show how they compare to the (presumed) counterexamples by reference to Dwyer's filtration on second homology. The development brings together many basic results on the nilpotent theory o f links. As a special case, a class of links only slightly smaller than "homotopically trivial links" is shown to have (free) slices on their Whitehead doubles.
Introduction In dimension four, the basic machinery of manifold theory, surgery and (5-dimensional) s-cobordism theorems, exist in the topological category when the fundamental group rr is "good" [FT] and is expected to fail for 7r free (and nonabelian) and in fact to fail for the "random" group. Nevertheless, even when rc is arbitrary many special surgery problems can profitably be solved. The theorem [F2] that the Whitehead double o f any boundary link is (freely) slice is an example. These applications all involve some representation of the surgery kernel by a submanifold M whose inclusion M C N into the source o f the surgery problem is 7h-null. Whereas all previous applications (IF2, F3, FQ] Chapter 6) required the second homology of M to be spherical, we find here (see Theorem 1.1 and Corollary 1.2) that the important condition is only that H2(M) = ~bo,(M), i.e. that the second homology lies in the co-term of the Dwyer [D] filtration as discussed in Sect. 2. This is an important philosophical point since for any n > 1, the "canonical" (or "atomic" compare [CF]) surgery The first author is supported by the IHES, the Guggenheim foundation and the NSF. The second author is supported by the IHES and the Humboldt foundation.
532
M.H. Freedman, P. Yeichner
p r o b l e m s - t o which all others restrict-can be chosen so that the kernel is carried by a ~l-null submanifold M with H2(M) = ~b,,(M). As elsewhere in this subject, taking the limit is the essential problem. On the way to the main theorem we develop in Sect. 2 the nilpotent theory of links and their (immersed) slices in a compact contractible 4-manifold, using only group theoretic methods (and not Massey products). This unified perspective contains many previous results (e.g. from [ M I , M 2 , D , T , K , L or C]) but uses only the largest possible indeterminancy for the invariants. A special case of our method shows in Sect. 3 that a class of links, larger than "boundary-links" and slightly smaller than "homotopically trivial links" have (free) slices bounding their Whitehead doubles (Theorem 3.1 ). This generalizes the main results of both [F2] and [F3] .
1. New surgery theorems
We describe a naive (map-less) surgery theorem and then its corollary in the formal setting of normal maps. Let N be a compact connected topological 4-maniibld, possibly with boundary. Let M C N be a connected codimension 0 submanifold with connected boundary. Assume that M is hi-null in N, i.e. the inclusion induces the zero map ~zj(M) -+ hi(N). Assume that H I ( ? M ) ~ HI(M). Then elementary calculations (see Sect. 3) show that H2(M) is tree. This says roughly that homologically M resembles a thickening o f a 2-complex. Note that the triviality of ~ I ( M ) ---+ zrl(N) implies a natural factorisatiou H2(M) --, 7r2(N) ~ H2(N) which we may use to define N + := NU/ff3-cells ) where the attachment is to the image in ~2(N) of a free basis [4 for H2(M). If [~1 and [~2 are two free bases they differ by a nonsingular integral linear transformation. Since any such transformation is a product o f elementary matrices there exists a "deformation" of the 3-cells realizing a (simple) homotopy equivalence N/~ ~- NI~. Thus N + is well defined. It has the same 2-skeleton as N and satisfies H2(N+; 2~[zrl ]) ~ H2(M) Q-~ Z[~zl ] ~ Hz(N; 2~[ni]). The nonsingularity of the intersection form on M (see Sect. 3) makes N + a Poincar6 space, but since Theorem 1.1 puts a manifold structure directly on N +, we will not offer a separate proof for this fact. With this notation, we state a naive surgery theorem for producing a manifold with the simple homotopy type o f N +. Sect. 2 treats Dwyer's [D] filtration of/42, n2 C ~,,, C ... ~h C ~b~,_l--- C q52 = / / 2 , appearing in the statement below. Theorem 1.1. I f the second homology of M
C N is in the co-term oJ Dwyer's .filtration, ~,,,(M) = H2(M ) then there exists a 4-man(fbM N', with
8N' = (3N = ~N + and a (simple) homotopy equivalence (rely?) ( N ' , t ? N ' ) ~ (N+,~N+), i.e. a manifold structure (rel~) on N +.
4-Manifold topology 11: Dwyer's filtration and surgery kernels
533
Now consider the formal setting of surgery. Suppose that N / ~ X is a degree 1 normal map from a topological manifold to a Poincar6 space X. There is the classical surgery obstruction 0 C L ] ' ) ( n I X ) to constructing a normal bordism to a (simple) homotopy equivalence N ' t ~ X . (We suppose here that if ~N+~/~ then .[[,w : ?N ~ ?X is already an equivalence and then the normal bordism mentioned above is required to be relative to the boundary.) It is always possible to normally bord f to a =l-isomorphism with K : = ker(H2(N; 2 ~ [ = I X ] ) ~ H2(X; g [ n l X ] ) ) a free ~[~zlX]-module so we assume that tiffs has been done. By definition, the surgery obstruction 0 vanishes if there is a (preferred) basis for the kernel K in which the intersection form is hyperbolic. We say that M C N represents the surgery kernel if He(M)c~)~ g [ n l X ] ~ H2(N;7Z[alX]) maps isomorphically to K. We have:
Corollary 1.2. Given M C N representin,q a stam&rd surgeo, kernel and sati,sfvin.g the hypotheses above: ~l-null, H I ( M ) ~ HI(PM), and q~,,,(M) = H2(M) then there is a normal hordisnt front J" : N -~ X to a (shnple) hontotopy equivalence .[" : N ~ ~ X. The essential improvement over Chapter 6 o f [FQ] is that to be useful for surgery a ~l-null 2-complex K representing the surgery kernel (a neighborhood o f K corresponds to M above) need not be spherical but only satisfy (/),,,(K) = H2(K). We now explain this condition in detail.
2. Group homology and the lower central series The lower central series of a group G is defined by G I := G , G ~+l : = [G,G k] for k > I and may be extended to all ordinals by defining G ~ : = ['-]/~<~ G[; for c~ a limit ordinal. We will be primarily interested in the cases k finite and the first limit ordinal ~. There is an equivalent geometric formulation in terms of maps o f .gropes.
Definition. A grope is a special pair (2-complex, cirele). A grope has" a class k = l, 2 . . . . . oo. For k = 1 a grope is d~fined to he the pair (circle, circle). For k = 2 a .grope is precisely a compact oriented smfaee Z with a sin qh, boundary component. For k .finite a k-~lrope is &'.fined inductively as Jbllow: Let {~,,fi,,i = 1. . . . . genus} be a standard symplectie basis oJ" circles f o r Z. For any positive integers Pi, q, with p, + q, > k and P,o + q,o = k f o r at least one index io, a k-grope is Jbrmed by .qluin~j p,-gropes to each ~, and qi-,qropes to each fi,. Finally, an oo-grope is a nested union q[" (k-,qropes, .fixed circle) .lbr all k > 1. The important information about the "branching" of a grope can be very well captured in a rooted tree as follows: For k = I this tree consists of a single vertex v0 which is the root. For k = 2 one adds 2 . g e n u s ( X ) edges to
534
M.H. Freedman, P. Teichner
v0 and may label the new vertices by eti,/3,. Inductively, one gets the tree for a k-grope which is obtained by attaching p,-gropes to e, and q,-gropes to /3, by identifying the roots of the p,-(resp, q,-)gropes with the vertices labeled by ~,(resp. fl,). The below figure should explain the correspondence between gropes and trees.
leaves
grope of class 5
associated tree root
Fig. 2.1.
Note that the vertices of the tree which are above the root t~0 come in pairs corresponding to the symplectic pairs of circles in a surface stage and that such rooted paired trees correspond bijectively to gropes. Under this bijection, the leaves (:= non-root 1-valent vertices) of the tree correspond to circles on the grope which freely generate its fundamental group. We will sometimes refer to these circles as the tips of the grope. The boundary of the first stage surface Z will be referred to as the bottom of the grope. Lemma 2.1. For a space X, a loop 7 E ~E(X) k, 1 <_ k < ~o, ( l a n d only i./'~ bounds a map o1" some k-grope. Moreover, the class q[ a grope (G, 7) is the maximal k such that 7 E ~zl (G) k.
Proof The first statement is proven by an induction on k, starting with the fact that the boundary circle 7 of a compact oriented surface Z with standard symplectic basis (<xi,fl,} is the product of commutators 7 = 1-[ [~i, fii]. Note that the "if-direction" is harder and uses the non-obvious re-bracketing fact [Gi, G q C_ G i+j, see e.g. IV, p. 27] For the second statement we only have to show that the boundary circle 7 of a k-grope (G, 7) does not lie in ~zl(G)~+l: Again this is best proven by an induction on k, starting with the fact that rtL(s is freely generated by all ~, and/3,. The Magnus expansion shows that 7 = l-I[cG/3,] does not lie in ~l(L') 3. Similarly, for k > 2, 7tI(G) is freely generated by those circles in a standard symplectic basis o f a surface stage in G to which nothing else is attached. Now assume that the k-grope (G,~,) is obtained by attaching p,-gropes G~, to c~i and qi-gropes Gfi, to fii, P~ + qi ~ k. By induction, ~, r rcl(G~,) p'+I
4-Manitbld topology I1: Dwyer's filtration and surgery kernels
535
and fl, ~ Tcl(G1~,)q' ~1 since p,,q, > 1. But the fi'ee generators of nl(G~,) and ~q(G/~,) are contained in the set of free generators of hi(G) and therefore ~, = fI[c~,,[I,] ~ zrl(G) ~+1. Again, this may be easiest seen by applying the Magnus expansion to hi(G), compare [MKS, Chapter 5]. [] Remark. Given 7 E ~Zl(X)~, one may write it as a product o f "right normed" commutators of the form [xl,[x2,[ .... xt] .... ]]. Such a commutator bounds a map of a very special k-grope, namely one whose rooted tree looks like
X2
Xk.1 /
X1
5p ."
\ root Fig. 2.2.
This implies that 7 bounds a map of a l/2-qrope of order k ductively defined to be obtained from a surface Z' by attaching order (k - 1 ) to a 1/2-symplectic basis {cr of Z'). This gives symmetric class of gropes. For each group G there is a least ordinal cr such that G ~ [~ > cr Call this stable stage G .... .
(which is inl/2-gropes of a particularly = G 1~ for all
Lemma 2.2. 7 r :zl(X) ..... ( / a n d only ([7 bounds a map o f an oc-#rope in X. Pro~[~ Follows from the definitions.
[]
Note. There are finitely presented groups with G 'n~ strictly smaller than G'". Geometrically, lying in G'" is equivalent to bounding maps of possibly unrelated k-gropes for each k < ~o. The lower central series is connected to homology and thus the rest of topology by Stallings' theorem and Dwyer's extension. Both o f these theorems are formal derivations of the 5-term exact sequence for groups: Given a short exact sequence I --* N --+ G ---~ Q---+ I
536
M.H. Freedman, P. Teichner
o f groups, the bottom part of the corresponding Lerray-Serre spectral sequence is an exact sequence of homology groups (with integral coefficients) H2(G) --~ Hz(Q) --~ N / [ N , G ] -+ H i ( G ) -~ HI(Q) ---+O. Stallings' Theorem [St]. / f a --~ ~ is a homomorphism ~71 groups induein,q an Lvomorphism on HI and an epimorphism on H2 then the imhu'ed maps a/cr h -~ ~/~h. are isomorphisms ./or all 1 <= k < (o. 1/'a ---+ ~ is an epimorphism then a/c;'" ---+7r/rc'" is also an isomorphism. For k => 2, let ~/~h(G) denote the kernel of H2(G)
--+
H R ( G / G h-I ).
Dwyer's Theorem IDI. Let cr --~ ~ imhwe an isomorphism on HI. Then ./or 2 < k < ~o the folhmqnq three conditions are equiralent: (1) f induces an epimorphism H2(cr)/~a(~)-+ H2(Tz)/~/~h(Tr). (2) f induces an Lsomorphism a/c~h -~ ~/~zh. (3) f induces an isomorphism H2 ( c~)/chh (a) --+ H2 ( Jr)~rich( 7r) and an injection H2((;)/r ---+ H2(rc)/@,+l(~). The previous two theorems apply directly to spaces (attaching cells of dimension > 3 to make spaces into K(z, l)'s does not affect these low dimensional statements) and we shall freely apply them in that context. For example, qbh(X) is defined to be the kernel of the composition H2(X) --+ Hz(K(rclX, 1)) : H2(lrI(X)) --* Hz(TzI(X)/~I(X) h 1). Then Dwyer's functor {/~h has a more geometric interpretation: Definition. A closed k-grope is a 2-complex ntade hy replacinq a 2-cell in S 2 with a k-~lrope.
Lemma 2.3. Dwyer's subst?ace qbh(X) o f H2(X) is precisely the subset q[ homoloqy classes represented hy maps q[" closed k-gropes into X. Proq[; Let rc : = r~L(X). We first observe that a homology class which is represented by a map of a closed k-grope is also represented by a map of a closed l/2-grope of order k. It is enough to show this for X a closed k-grope and the homology class the generator of H2(X) 9 . Cut out a 2-cell from the bottom stage o f X. Then the boundary ? of this 2-cell lies in nl(X\2-cell) h. Now use the remark after Lemma 2.1 to get a map o f a l/2-grope of order k bounding 3' and reglue the 2-cell to it. To see that a homology class of a space X represented by a closed 1/2-grope of order k lies in q~h(X), take a representative bottom surface Z of the 1/2-grope o f order k. The l/2-symplectic basis o f curves o f Z' to which the next grope-stages attach are mapped to ~h-I and so are trivial under the inclusion K0r, 1 ) C K(~/Tt k-I, 1). This means that the homology class of-Y' becomes spherical and hence trivial in K(Tr/~ k-l, 1). Now consider a surface Z' mapped into X which is null homologous in K(~/~ h-I, 1). Let the null homology be represented by a map F o f a 3-manifold W. Think o f K(~/Ir h-I, 1) as X U2-cells U 3-cells U . . . or by thickening as X U2-handles
4-Manifold topology I1: Dwyer's filtration and surgery kernels
537
O 3-handles O .... Make F transverse to the ascending (singular) manifold A of the attached handlebody. The compact 3-manifold W\, ~ ( F I(A)) is a bordism in X between Z' and another map of a surface .[ : Z --+ X which, by inspection, has the property that a I/2-symplectic-basis of Z maps to ~, (core 2-handle), i.e. to the relations ~z~-I It follows that (Z,.f) (which is homologous to Z') extends to a map of a 1/2-grope of order k into X .
[Zi
~N (F-I(A)) with half-basis of cocores marked
Fig. 2.3. John Milnor [M2] defined certain numerical invariants ~L for a link L in S s which extend without difficulty to the case of a link in an integral homology 3-sphere Z. We will define these invariants after recording some relations between the size of O~(Z\L) and the comparison of the groups ~I(Z\L) and ~l(~~ with the free group on n generators. The symbol //'~ denotes 0-framed surgery on an n-component link L in Z. Let V C Z\L denote a bouquet of meridians to L and set F := ~I(V). Lemma 2.4. In the above setting, the,fidhm'ing statements are equivalent.fi, r k_>2: (i) All (untn,isted) longitudes of L hound maps of k-gropes in Z\L. (ii) H2(Z\L) = Ot+l(Z\L). (iii) The hu'lusion V C Z\L induces an isomoq~hism
F / F ~+l ~---~,( Z \ L ) / ~ t ( Z \ L ) h+' (iv) H2(,U~
= r176
~- E," ). (v) The inclusion V C / f ~ F / F ~~ l
(J0r k = 2 this should he read as H2(Sf~
induces an isomorphism (5/'~
(~'~ L ))~ 9
538
M.H. Freedman, P. Teichner
Proof (i) =~ (ii) By Alexander duality, H2(X\L) is generated by the n tori parallel to the components of L. These are obviously in ~/~+t(X\L) if the longitudes bound maps of k-gropes. (ii) r (iii) follows directly from Dwyer's theorem since H 2 ( V ) = O. (iii) =~ (i) Using (iii) we may write any (untwisted) longitude I as 1 ~ F / F ~+l. If m E F is the corresponding meridian, the relation [m, l] = 1 in the link complement becomes [m, 1] = I in F / F ~+1. Using the Magnus expansion [MKS, Chapter 5] (which is explained in detail below) this implies that l E F h and by (iii) 1 E nl(XkL) k implying (i). (i) =~ (iv) Since k > 2, the longitudes of L bound surfaces in X\L. These can be capped off by the cores of the 2-handles in 5f~ to get a basis for H2( 5/~~ L ) ) ~- 2~". This construction shows that H2( 5/;~ L ) ) = ~/~ ( 5/~~ L ) ) if the longitudes bound maps of k-gropes. (iv) r (v) is again Dwyer's theorem. (v) =~ (i) the commutative triangle F
~
t, n I ( Z \ L)
n I (~'(L)) leads to three isomorphisms if one divides by the k th stage of the lower central series: This is true by assumption for [] and therefore becomes injective. Moreover, ~ also becomes surjective in any nilpotent quotient because the meridians become normal gener'ators since HI(Z') = 0. But in any nilpotent group normal generators are also generators which can be proved by an induction on the nilpotency class using the fact that a = b mod N implies x ~ - x b mod [G,N] for any elements a,b,x in a group G , N C G. This proves that (v) implies an isomorphism
i. : n I ( X \ L ) / g I ( X \ L ) ~ ~- , n r ( 5 / J ~ 1 7 6
k
from which (i) follows since the longitudes become trivial in n l ( S ~ 89 We may now define weak (with large indeterminancy) ~-invariants of an oriented link L C X inductively as follows: The ?7-invariants of length 1 are defined to be zero. Assume that statement (iii) of Lemma 2.4. holds for some k => I. We define integral valued ~L-invariants of length (k + 1) using the isomorphism from (iii) to get well defined elements {/ E F / F k+j from the (untwisted) longitudes of L. Then the Magnus expansion (given by t/Ti v-~ 1 +xi)
M : F = F(mj . . . . . m,,) ~ 7Z{xl . . . . . x,,}
4-Manifold topology I1: Dwyer's filtration and surgery kernels
539
into the (units in the) ring of non-commuting power series can be used to define the numbers fiL(l,j) via ~ T L ( I , j ) x / := m ( / j ) . 1
Here 1 runs through all possible multi-indices but only those with exactly k entries (leading to the invariants ~L(I,./) of length ( k + 1 )) are interesting: This follows from the fact thai the Magnus expansion maps F' to power series of the form I + terms of degree >= i (all x, have degree 1). By assumption // E F a / F a+l and thus exactly the terms of degree k in M ( / / ) are well defined and interesting. One knows [MKS, Chapter 5] that the Magnus expansion is injective and that the associated graded map (given by a H M ( a ) - - 1) Gr(M)" G r ( Y ) : = (~DFa/F h+' ---+ (~degree k/degree (k + 1 ) = : A,, /,>1
/,~1
into the free associative ring A,, in xl .... ,x,, is a Lie algebra isomorphism from Gr(F) (with its Lie bracket induced by the group commutator [a,b] = aba tb t) onto the (free) Lie algebra inside A,, (with Lie bracket [0~,fl] = ~fl fic~) generated by xl ..... x,,. This implies that the ~TlTinvariants of length (k + 1) vanish if and only if the (equivalent) conditions of Lemma 2.4. hold for k + 1. Moreover, it implies that the ~7L-invariants satisfy certain relations (which Milnor called shuffle symmetries) to keep G r ( M ) ( / / ) = M ( / / ) - 1 in the Lie algebra on the xi. -
Lemma 2.5. The fiL-invariants are cyclically symmetric, i.e. ~L(il . . . . . ix,j) = -fiL(j il . . . . . i~ ) if all ~L-invariants o f len,qth < k vanish. Remark. For k = 1 this is the well known symmetry of linking numbers since one easily checks that }TL(i,.j) is the linking number between the i-th and j-th component of L. P r o o f ( o f cyclic symmetry). The longitudes of L give elements /j E F t / F k+l which lead to elements [ml,/j] C FX+l/Fk+2. The 5-term exact sequence for the extension 1 ~ F x+l ---+F --+ F / F x+j ---+ l gives an isomorphism H 2 ( F / F k+l ) -~ F k + l / F ~+2 which we compose with the isomorphism of statement (iii) to get //2( 7I" 1 ( X \ L ) /
R" I
(Z'\L)/'+1) ~z Fk+t / F k+2
.
It is easy to check that the elements [mj,{i] correspond (under this isomorphism) to the image of the tori Ti C S \ L parallel to the components of L
540
M.H. Freedman, P. Teichner
under the natural maps H2(Z'\L) --+ H2(TcI(L'\L)) ---+Hz( ~ZI(L'\L )/~zt( L~\L )~+1 ) . n
The cyclic symmetry of the FL can now be derived from the relation ~ / = t T/ = 0 in H2(X\L) as follows: mapping this relation to F t + l / F t+2 and then applying the graded Magnus expansion gives the following relation in A,,:
0 = ~x/Gr(M)(//) i=J
- G r ( M ) ( / i ) x , = ~ E fi,,(l,j)(x/xt - x , x / ) . i=l I/1=~
Focusing on the coefficients at xzx/ for some fixed index 1 = (i, . . . . . i~) one immediately recovers the relations
ilL(it . . . . . ih,j) = ~L(,/, it ..... i~ ).
E5
Let Z be the unique contractible 4-manifold with boundary S, see IF1]. We will say that a link L C 2; is 4D-homotopieally trivial if it extends to maps A, : D 2 ~ Z with disjoint images.
Remark. For (Z,S) = (D4, S 3) this notion agrees with Milnor's [M1] "link homotopy" as we shall prove below. But if S is not simply connected then there are knots in S which are not null-homotopic but they bound a map A : D 2 - + Z since ~ ( Z ) = {l}. Definition. Let a group G be normally generated by elements gi, i ~ I. Define the Milnor group o f G (rel g,) to he the quotient MG := G/[.q[,, .q],,] = 1 Vxi, y, E G, i E l . In [FT, Sect. 3] we show that for Ill = k the Milnor group M G is a finitely presented nilpotent group of class < k, see [M1] in the case o f link groups. In particular, M G is a quotient of the free Milnor group
mFk : = F(mL . . . . . m~ )/[m~,,m~.,]=l . Usually the normal generators are clear from the context, for example M n l ( S 3 \ L ) or M ~ I ( Z \ A ) always refer to the meridians (to L, resp. Ai) as normal generators.
Lemma 2.6. Let L C X be 4D-null homotopic and let A = (A, . . . . . A~) be a null homotopy. Then the meridian map induces an isomorphism MFk ~ M g l ( Z \ A ) . P r o o f The second homology of Z \ A , if A is in general position, is generated by the "Clifford tori" linking the transverse double points o f A. As in [FT, Corollary 3.2], working modulo any term N of the lower central series, a
4-Manitbld topology ll: Dwyer's filtration and surgery kernels
541
bouquet of meridians W to L induces an epimorphism 0:
nj(W ) / ~j(W) N
~)
g~(W ~ 2-cells) / 7r~( W ~ 2-cells)
~, n f ( Z \ A ) / ~ z I ( Z \ A ) N
N
The "'meridians" and "longitudes" of the Clifford tori now lilt along 0 to conjugates of the basic meridians of W, m~ and m(~. The 2-cells in the above diagram are attached to the commutators [m,~,ml~] and the map on space level extends to an epimorphism on//2. By Stallings' theorem, 0' is an isomorphism while 7, by the nature of the relation 2-cells, induces an isomorphism on Milnor groups. Since we may assume N larger than the nilpotency class = k of the Milnor group MF;,, 0 induces an isomorphism MFI,. ~ M n I ( Z \ A ) . [] R e m a r k . Note that M n I ( Z \ A ) is obtained from n l ( Z \ A ) by adding finitely many relations of the form Imp, m,] = 1 and these may be realized by introducing additional self-fingermoves to the A~. Therefore we may always assume that the null homotopy A satisfies M~r(Z\A) ~ 7rl(Z\3).
In [M1] Milnor showed that the Magnus expansion induces a well defined (and still injective!) homomorphism M M : MFt -+ Rh
into the (units of the) ring R~ which is the quotient of the free associative ring A~ by the ideal generated by the monomials x,~...x,, with one index occuring at least twice. If l~+l C Z\L is an additional component in the complement of a 4D-null homotopic link, L + := L U lh+l, then we define the /l-invariants of L + by the equation MM(I,~ ~l ) : ~ ItL +( l , k + 1 )x,
E R;,
I
where I is a multi-index with nonrepeating entries from {1 . . . . . k} and /~.+] E MF,,, is obtained via the isomorphism in Lemma 2.6. (and the inclusion Z'\L --~ Z \ A ) . Using the injectivity of M M and the remark after lemma 2.6 we conclude
that L + is 4D-homotopically trivial if and only if all lit +-invariants vanish. But Milnor showed in [MI] that this is also true for his notion of homotopy for links in S 3. Therefore, 4D-homotopy and link homotopy agree in this case. The commutative diagram M
nl (Z'\L)
~
F
Mnl(Z\A)
~
MF
~ ~{xi} ~
R
542
M.H. Freedman, P. Teichner
shows that t~(I) = ~(1) if both invariants are defined, in particular 1 must have nonrepeating indices. L e m m a 2.7. For a link L C S and k > 2 the Jbllowiny two statements are equivalent to all statements in 2.4. (a) All ficinvariants o f lenqth < k vanish. (b) IJ L is any link oJ k (or fewer) components made from untwisted parallels ~]" L, then L is 4D-homotopically trivial. Proof The equivalence of (a) with the statements in 2.4. follows from the injectivity of the Magnus expansion as discussed above. (a) ~ (b) First note that if the longitudes of L bound maps o f k-gropes in Z'\L then the longitudes of L bound maps of k-gropes in S \ L (and vice versa if L contains each component of L). By Lemma 2.4, the ~L-invariants of length < k vanish. Choosing an ordering of the components of L we can now inductively prove that also all the llL-invariants vanish and thus L is 4D-homotopically trivial. (b) => (a) By a simple induction on [I I we may assume that all ~Linvariants of length < n < k vanish and thus we can use our definition for g o ( l ) for III = n + l. To prove that the invariant ~L(I) is trivial choose the link L = L(I) to be formed from ni parallels of L, where n, is the number of times the index i occurs in 1. Also form a new multi-index )" from 1 by replacing the n, occurrences of the index i with distinct indices i/i,...,i/n,. By definition, )" has nonrepeating indices and labels the (n + 1 )-component link L. Therefore, the invariant ILL(I ) vanishes since we are assuming (b). It follows that the invariant ~TL(I") vanishes. But a straightforward checking of definitions shows that this invariant equals gL(1). One only has to observe that the relabelling of L to L changes a meridian m, to the product mi/i...mim ' and thus xi = GR(M)(mi) is replaced by the sum xij~ + ... + xij,, . [] For the final construction in Sect. 5 we need one more Lemma on links in a homology 3-sphere S which uses the cyclic symmetry of fi-invariants in this setting. L e m m a 2.8. Let L C S have vanishin,q ~-invariants o f lenyth < k and let R p C S \ L be a link with each component lyin9 in ~I(S\L) k. Then there is a "weak" homotopy (not a Milnor link homotopy avoiding certain collision but just a general I-parameter motion in S\L ) o f R / to R so that the link L U R has vanishin,q ~-invariants o f len(Ith < k. Proof Each component r/ of R' bounds a singular grope of class k in S\L. By general position these gropes have disjointly imbedded 1-spines S,. The desired weak homotopy pushes each r~ to a neighborhood of Si where it bounds an imbedded grope Gi of class k, Set ri := OGi. Since the various Gi a r e disjoint and in particular Gi N (r i U L) = ~ for i:~j, we see that the longitude of ri lies in ~l (X\LUR) k so all ~-invariants of LUR of length < k ending in an
4-Manifold topology 11: Dwyer's filtration and surgery kernels
543
R-index vanish. By cyclic symmetry of the ~ (Lemma 2.5), the only ~-invariants of L U R of length < k which could be nonzero would be those with only L indices. But the ~-invariants are natural under filling in the R-components so this would imply a nonzero fi-invariant of length < k for the original link L against our hypothesis.
3. Whitehead doubles of links
To introduce the construction used to prove Theorem l.l we present a much simpler application which generalizes the main theorems of IF2] and IF3]. Let L C S 3 be a smooth link. Wh(L) denotes an unramified, untwisted Whitehead double. This means each component /, has been replaced within its neighborhood by a component Wh(/,) of either form
or
Fig. 3.1.
so that the Seifert form on the punctured toms T, bounding Wh(/,) within a neighborhood o f / , is of the form I~ g I" This last condition makes "untwisted" precise. The • ambiguity in the choice of the clasp in Fig. 3.1 relates as we will soon see to the sign of a double point in 4 dimensions. This sign will not be relevant in our discussion so the term "Whitehead double" refers to any of the 2 #~Lt possible links. In [F2] it is shown that if L is a boundary link then Wh(L) is slice (i.e. bounds disjoint flat disks in D 4) with rq(D4\slices) freely generated by meridians ("free slice"). In [F3] it was shown that the same conclusion applied to exactly those two component links with trivial linking number. Definition. A smooth link L = (/I . . . . . #,) in S 3 is (homotopically trivial) + i[" the n links o f (n + 1 )-components obtained by adding a parallel copy o f a single c o m p o n e n t / , , i = 1. . . . . n are each homotopically trivial in the sense o f [MI] (or 4D-homotopically trivial as in Sect. 2). Theorem 3.1. I l L C 3 3 is (homotopically trivial) + then Wh(L) is freely slice. Remark. It is unknown whether homotopically trivial is an adequate hypothesis for this theorem. The still weaker hypothesis that all linking numbers o f L vanish would suffice if surgery "worked" for free groups. We note that (homotopically trivial) + is equivalent to the vanishing of all ~-invariants with at
544
M.H. Freedman, P. Teichner
most a single repetition of one index. It is also equivalent to the meridian map inducing an isomorphism of Milnor groups M F ~ M~I(.~P~ L e m m a 3.2. Let L = (/i . . . . . /,,) and L' = (t'1 . . . . . t~) he a parallel copy. The condition L is (homotopically trivial) + is equivalent to the existence ~[" disks {Ai} U {A;}, 1 < i < n, properly mapping into B 4 with ?A, = /,,~A~ = /; and di.sjointness assumptions: A, N A I = ~ /~r i + j and A~ A A I = ~ f o r all I <=i, j <= n. P r o o f The condition L homotopically trivial is known (see [L] or [FT, Lemma 3.3]) to be equivalent to the existence of disjoint {A,, _< i < n} as above. In these terms, the + condition means that for each _-< j ~ n the {A/} can be chosen so that [/~] = e C gl(B4\{A~, ] =< i ~ n}). However, by Lemma 2.6 these groups for different choices of {Aj} all have a common quotient ME,, the free Milnor group on n generators which is itself realized as ~I(B4\{A, 1 < i < n}) for sufficiently complicated choices of A,. Thus the null homotopies {A~, 1 < i _< n} all exist disjoint from {A,, I _< i _< n}. CA With the notation below the pair (B4;. ~"(hl )LS.,'| '(h2)) is a concrete model of a (positive) plumbed pair of 2-handles, see [FT]. The notation is: hi and h2 are two H o p f circles in S 3 : ~'?,B4 C I~ 2 and the ,t"'s are 3-dimensional solid torus neighborhoods of these. All orientations are taken standard with respect to complex multiplication. Reversing the orientation along one Hopf circle gives a negative plumbed pair. More generally, handles may be plumbed + or - together in many (disjoint) places and self-plumbed to produce the kinky handles of [F1 ]. L e m m a 3.3. The effect o f introducing a 9 plumbing (or s e l f plumbing) on the underlyin.q (Kirby) handle diagram o f a handlebody is to introduce a new 1-handle and a i clasp ~[" the attaching curve(s) o f the 2-handle(s) being plumbed over this 1-handle as shown below
i i
after (+ case)
before Fig. 3.2.
, 0,
after (- case)
4-Manilbld topology
I1: Dwyer's
liltration and surgery kernels
545
Proof (sketch). First identify a l-handle in the plumbed handlebody by taking a neighborhood of two arcs leaving the "origin" o f the plumbing on the two core sheets. The new attaching curves for two-handles are as before except that the attaching curves induced in the plumbing must now run up this 1-handle, i.e. through the dotted circle, in the diagram, clasp, and return. The sign o f the clasp is worked out from the Hopf link model introduced above. [5
o/" Theorem 3.1. We introduce a specific 4-manifold N whose boundary ?N is 0-framed surgery on Wh(L),?N ~ / f ~ Set No : = B4ULUL, 2-handles, the result of attaching 2n 2-handles with framing = 0 to the link L and its parallel copy L'. Now set N := N0/plumbings where for each 1 _< i _< n one plumbing is introduced between the 2-handles attached to /, and /~. The sign of the plumbing is, for each i, chosen to agree with the sign o f the Whitehead doubling of the i-th component. Pror
Lemma 3.4. ?N ~ / / ' ~ n'ith the isomorphism cartTinq the meridians to the l-hamlles (see Lemma 3.3) to the meridians to Wh(L).
Proof Inside a solid torus we have the following Kirby calculus identity:
I axis I I
// / /
//
\
/~
X
\ \ ~ a•
isotopy
o&
/ axis --
kJ
"~(~_ on3) / / ~ ~
II ~is~176 f f
~ onO ~
" --
\ axis
\\\
I -kj"
I
Fig. 3.3.
Note that the z-axis initially marks the complement of the solid torus. All clasps have been drawn ambiguously to imply the uniform treatment o f both cases. Now apply this identity simultaneously in the n solid torus neighborhoods of {/i} to finish the proof. IS] Next we cap off the cores of the 2n plumbed handles with the disks
{Ai, A;, 1 < i <= n} produced by Lemma 3.2 to obtain an immersed union S of 2n spheres in N. (In analogy with Theorem 1.1 we would call the closed
546
M.H. Freedman, P. Teichner
regular neighborhood of these spheres M but in this simple case the homology of M is spherical and we may finish directly). The fundamental group of N is freely generated by the n plumbings, that is, the solid torus neighborhoods ,.V(/i), 1 < i < n give a disjoint collection of 3-manifolds which by the Pontrjagin-Thom construction determine a map to ~/','=l $1j which classifies the fundamental group of N. It is easily seen that the disjointness conditions of Lemma 3.2 assure that every loop in S reads the trivial word as it crosses the various .iV(~,). Thus S is Tel-null. Also the intersection form carried by S is (~,1 ~ 0 I" The plumbings contribute the nontrivial entries; the double points ! d~ (1 dj contribute nothing since link(/~,{}) = 0 Vi, j. While it is unknown whether S is homotopic to a disjoint collection of imbedded hyperbolic pairs, it is shown in [FQ, Chapter 6] that such hi-null collections of spheres are s-cobordant to disjointly imbedded hyperbolic pairs. This is adequate to "complete surgery", that is to produce a 4-manifold N ' with (3N' = 0N and with a map 0 to a wedge of circles N ' ~ SiI which is an isomorphism on ~l and with the meridians to the l-handle curves for the diagram of ~3N = ~')N~ mapping degree 1 to the n circle factors. Clearly H2(N I) = 0 and therefore by [FQ, 11.6C(I )] 0 is a homotopy equivalence. It is now standard to observe that S 3 x 1 Uwh~L)x I (framed 2-handles) tO N ~ ~ B 4 where the last union uses the identification of Lemma 3.4 and (?N ~ = ~3N. The 0-framed 2-handles now extend through the product collar S 3 x 1 to become the desired free slices on Wh(L) x O.
4. A special case This section contains the proof o f Theorem 1.1 in the presence of the extra assumption that nI(N\M) --, rq(N) is an isomorphism. This is often described by saying M is nj-negligihle and is a very familiar condition in 4-manifold theory. Removing this assumption adds a final layer o f subtlety to the argument which will be defered until Sect. 5. But let us first collect the elementary homological consequences of the other assumption, namely that HI(~?M) ~ HI(M) is an isomorphism. Considering the horn-dual of the isomorphism and the universal coefficient theorem, we see that HI(M) --* HI((?M) is an isomorphism. By Lefschetz duality H3(M, (~M) --~ H2((?M) is also an isomorphism. From the exact sequence of the pair (M,(?M) we now conclude that H2(M) --~ H2(M,~3M) is an isomorphism and HI(M, OM) = 0. Similarly,
0 = Ht(M,(?M)
TM
H 3 ( M ) ~- free(H3(M)) | torsion(Hz(M)),
so H2(M) is free and H2(M) ~- H2(M,(?M)
TM
H 2 ( M ) ~ free(H2(m)) | torsion(HI(M)),
and therefore HI(M) is also free.
4-Manifold topology I1: Dwyer's filtration and surgery kernels
547
Definition. We say a group n is weakly-para-free (f there is a map .fi'om a free group F --+ n inducing isomorphism on all the .finite quotients o]" the lower central series F/F ~ ~ n/n ~, k = 2, 3 .....
We know that H I ( M ) ~ H I ( ? M ) are free abelian so let ~: " V ~ ?M be any map from the wedge of n-circles to ?M inducing an isomorphism on H~. Now considering s:# " nn(V) ~ nl(~'~M) one finds Lemma 4.1. n l ( ? M ) and z h ( M ) are both weakly-para-free with the inclusion q[" the wedge V inducing the required map. Also the map n I ( ? M ) --~ h i ( M ) is an isomorphism modulo an)' .finite term q[' the lower central series. P r o o f Consider the composition V & i~M 2+ M. Since the homology of M is "nearly spherical" in the sense that 4),,,M = H2(M), Dwyer's theorem tells us that both maps i and i o ~: with target M (which we already known induce isomorphisms on HL) induce isomorphisms on all finite quotients of the lower central series. Applying the functors nl/(nl)~ k > 2 to the diagram proves the Lemma.
Define MI C N to be
MT := M U 2-handles/plumbings and self-plumbings. {;',}
(4.1)
The submanifold MI is simply what can be produced from M by using the hi-null and ltj-negligible hypotheses on M. Specifically, take any collection of n disjointly imbedded circles {7'1. . . . . 7,} in ~M homotopic to the petals of V and cap these off by n general position null homotopies 61 . . . . . ~, whose interiors are disjoint from M. The submanifold Mi is simply a regular neighborhood of M U (U;'_16,). By the basic theory of topological immersions [FQ, Chapter 8] Mi may be described combinatorially as in line (4.1) where the n 2-handles determine definite normal framings f , on 7, ()'i and its parallel 3,~ should bound chains with intersection number 0 in the plumbed 2-handles). Let X := .~/~cM((Ti,J)), i = l . . . . . n) be the abstract homology sphere which results as the boundary o f the abstract 4-manifold MU{:.,,/, / (n 2-handles). It is abstract in the sense that it is not a submanifold o f N, but our strategy is to construct another abstract manifold M2 with a canonical isomorphism c3M2 ~ ?M1 and such that H2(M2) is represented by a hi-null collection of spheres. To begin the construction of M2, let Z be the unique contractible manifold bounded by S, see [F1]. We may consider (Kirby) handle diagrams drawn in S = (?Z. To start notice the n-component link (ml . . . . . m , ) C S consisting of meridians to (3'l . . . . . y,,) C 0M. O-framed surgery on (ml . . . . . m , ) is naturally identified with ~?M, reversing the initial surgery. Furthermore, the 0-framed meridians /{I,.-.,/~,~ to m j . . . . . m, become (TJ,.J']) ..... (~/n, J , ) under this identification. Now Lemma 3.3 can be exploited to give a Kirby diagram for 8Mi
M.H. Freedman, P. Teichner
548
in 2; as shown below
17
\m,
Fig. 4.1.
What we see are the meridians m,, their meridians It~ modified to It, by clasps (induced by some number of (self-)plumbings and 1-handle curves linking these (self-)clasps. The figure shows ttr and tt,, with one clasp and one, respectively two, selfclasps. The geometric shape of the m, reminds us that their detailed position in 2," is unknown. As in Sect. 3 the sign :t: of the clasp is intentionally suppressed in the figure. We define 1142 as realizing the boundary equivalent link diagram in ~Z = X where each clasp has been "blown up" to a 0-framed Hopf link and each iti, 1 <- i <_ n, has been made into a I-handle (given a dot) as shown (locally) below.
Pi
,(
Fig. 4.2.
Thus M2 is defined as Z U l-handles U 2-handles according to Fig. 4.1, as modified locally in Fig 4.2. We may change the handlebody structure of M2 (but not its homeomorphism type) by Morse cancelling each It, with mi, 1 -< i -< n. The result (using the same multiplicities as in Fig. 4.1)
4-Manifold topology I1: Dwyer's filtration and surgery kernels
549
Fig. 4.3.
Lemma 4.2. Any link consis'ting o/ untwis'ted parallel copies o / ' L
:=
{ml ..... m.} in S bounds di,~johtt maps o[" disks into Z. Proof According to Lemmas 2.4 and 2.7 exactly the links L C S with nl(S\L) weakly-para-fi'ee admit disjoint maps of disks on any number of parallels. Since ?/~?(L) = ?M, Lemma 4.1 says that nl(,c/~)(L)) is weakly-para-free. By Lemma 2.4 this is equivalent to nl(Z\L) being weakly-para-fi-ee. [] Using the 2-handles in Fig. 4.3 to form the "northern hemisphere" and the singular disks in Z located by Lemma 4.2 as "southern hemisphere" we see that the entire second homology of M2 (with group-ring coefficients) is freely generated by a nt-null collection of spheres with hyperbolic intersection form. The verification is much the same as in Sect. 3. As in Sect. 3 we use [FQ, Chapter 6] to s-cobord the spheres to a disjointly imbedded collection of hyperbolic pairs. Removing these pairs by surgery yields a 4-manifold M 3 with ~ M 3 = ('~M2 = ?ML, and as at the end of Chapter 3 a
M3hw
homotopy equivalence t o a wedge of circles which takes each meridian to a 1-handle of Fig. 4.3 degree I to a distinct circle factor. The free generators correspond to the double points of the null homotopies {61 ..... 6n}. The manifold N: asserted by Theorem 1.1 is defined as (N\MI)Uid M3We now construct a map g : N + --+ N'. Set gtN\Mt : = identity. Near each self-plumbing of the 2-handles in the combinatorial description (4.1) of Mi we may locate a solid torus dual to the arc which changes sheets at the selfplumbing. These tori, by the Pontrjagin-Thom construction, determine a map MI --~ W which extends (uniquely up to homotopy) to a map MI U3-ce~ls --+ W. On ~?M1 = (?M3 this map restricts to h and thus provides an extension q
of the identity on (')M~ = ~ M 3 tO a map Mt U 3-cells ~ M 3 . Let this last map be g restricted to ML U 3-cells.
550
M.H. Freedman, P. Teichner
To check that g induces an isomorphism on gl we ignore the 3-cells and consider two interconnected pushout diagrams of groups.
gI(MO = flee
Z
rcl(N' )
nJ(MI)
T,, r c ' ~ T r L ( N )
n~(N \ Mr) The groups hi(N) and nj(N ~) are pushouts of maps (1,2) and (3, 2) respectively. Map 4 is the restriction of q and map 5 is any splitting of map 4. By construction, map 6 is the restriction of g and we want to show that map 7 is its inverse. By the pushout property map 4 (resp. map 5) induces map 6 (resp. map 7). Since map 4 o map 5 = id, map 6 o map 7 = id on nj(NI). To check that map 7 o map 6 is also the identity we need to know: ker(map 4) C ker(map 8)
(4.2)
for map 7 o map 6 would then change a standard-form word for h i ( N ) only on the letters in nl(Mt) and these letters would only change by an element of ker(map 5 o map 4) which is no change at all in hi(N). To check (4.2) observe that ker(map 4) = normal closure (image map 9) and that nt-nullity states that map 10 is zero. With n := nj(N +) ~ nj(N'), define K, := ker(Hi(N; ]g[n]) + / / i ( N ' ; 2g[n])),
K i := coker(Hi(N',~7[n])
~ H'(N,~[n])).
Since N -~ N' is a degree l map cap product with the fundamental class induces isomorphisms K, ~ K 4-i, i 0 , . . . , 4 . Also since H I ( 9 ;~[n]) is the first cohomology of the n-cover with compact supports it depends only on g implying K3 ~ K I = 0. Thus /s is the only non-trivial h o r o l o g y kernel. Set Mz C N to be the inverse image of Mi. Then M] consists o f M = (n-copies of M ) union various singular disks A with H2(A,~?) --~ HI(M) an isomorphism. =
Thus H2(M)~H2(I~I ) is an isomorphism. The image o f / ~ t under ,q is some cover of M3 -~ W and so has no second homology. Excision then implies K2 = image(H2(/141 ) ~ H2(N) ~ Hz(N; 7/[gr])),
(4.3)
H2(~II ) ~ H2(~4 ) ~- H2(M) @z 7Z[n].
(4.4)
and also
4-Manifold topology I1: Dwyer's filtration and surgery kernels
551
According to the beginning of Sect.4, our hypothesis HI (?M) ~ HI (M) implies that H2((?M)---, H2(M) is zero and by the Mayer-Vietoris exact sequence the map H 2 ( M ) ~ 7/[n] --+ H2(N) induced by the inclusion must be an injection. It follows from lines (4.3) and (4.4) that 1s ~-H2(M)c~)z;gn. Precisely a free basis for this module is killed by 3-cells in passing to N +. It follows that
g : N ~- --~ N' induces an isomorphism on Hi(;2~[n]), i -- 0 . . . . . 4 and by Whitehead's Theorem .q is a homotopy equivalence. Since the only interesting part of ,q is a 2gnL(M3)-homology equivalence MI U 3-cells-+ M3 any torsion would come from Wh(nl(M3)) = Wh(free) = 0. Thus ,q is a simple equivalence. This completes the proof of Theorem 1.1 under the assumption that M is hi-negligible.
5. The proof of Theorem 1.1 This section completes the proof o f Theorem 1.1. Let 6' = {6'I . . . . . 6,',} be null homotopies in N for 7J .... ,Y,, which initially leave M in an outward normal direction but may return to M. Since Hl(c~M) -~ HI(M) is injective each component C; o f 6 ' - J ( M ) may be replaced with an orientable surface C, r = (?C; and C mapping to ~3M, extending 6' on ?C'. Set/5 = (/5'\UC)U(UC) together with the map to N. By construction, H I ( U C ) --, H1(/5) is an epimorphism so a symplectic basis (c~,fl) for Hi((5) may be chosen to consist of imbedded hyperbolic pairs o f loops in UC; by hi-nullity the loops bound singular disks # in N. Thus we have capped off {7,} with capped surfaces /5 U ~:t Putting things in general position we have i5 A M = (~6 and ~:'N M = some planar surfaces in M. If nI(M,(?M) was trivial we could homotope t:' to achieve that t?' N M is a disjoint union of disks, an advantageous condition. By Lemma 4.1 we do know however, that for all k
nt(OM)/nl(~')M) ~ --~ n t ( M ) / n l ( M ) h is an isomorphism. Let c denote the number of components of ~J N M. We fix a large number K, to be specified later, and add tubes along ~')M to ~;~, to form ::, so that ~ NPM consists of a collection of c circles which lie in n~(~')M)K and so that ~:\M is a collection o f disks with a total of c punctures. It is now possible to draw a handle diagram relative to some contractible manifold Z, as in Fig. 4.1 to describe the boundary of a neighborhood of (M U/5 U ~:). Some "small" changes made to /5 and ~: improve the diagram to the type drawn below. After these changes we think (roughly) of the diagram as representing a 4-manifold M2, although we have yet to interpret the dotted components lq .... ,ll,, which may not form a slice link. If these components are not a slice link in Z the diagram only makes sense as the description of a 3-manifold. Later we will arrange that these components are slice in a space derived from Z, namely in Z # S 2 x S2's, allowing a 4-dimensional interpretation of these components as "pseudo-l-handles" in a stabilized diagram for a manifold
552
M.H. Freedman, P. Teichner
M +. Below we have the analogue of Fig.4.1 as modified by Fig.4.2. The background space for Fig. 5.1 is the homology sphere Z" = (')Z = ,9",'M{~,'l. . . . . 7,}-
0
0
i
-
Fig. 5.1.
The new feature in Fig. 5.1 is the "tumor" on the left. It arises (along with many similar copies on all ~t, which have not been drawn) by cancelling the "Bing pairs", associated in the diagram with the surface stage 6, with the 2-handles corresponding to ~:. The punctures in ~: give rise to the new 1-handles and "rectangular" components as in Fig. 5.2 below.
~\Mwith onepuncture~ ~~l"~J"')L__._~~\
......
. . . . . . . .
t
~
with2 punctures
I Fig. 5.2.
Absent (1) (2) (3)
from Fig. 5.1 are representations of: (~, 6) intersections, (~:,t:) (self)-intersections, and nonzero e-framings.
4-Manitbld topology I1: Dwyer's filtration and surgery kernels
553
By "spinning" (see [FQ, 1.3]) the framings of the 2-handles in Fig.5.2 may be made zero. This introduces (new) intersections of type (1). All intersections of type (1) and (2) can be removed by a move in which a sheet of 6 containing a bit of ?4: is pushed along an arc in c through the troublesome intersection point (see [FQ, 2.5]). The cost of this move is additional (6,~i)-intersection points but we have allowed for these in Fig. 5.1. A final improvement (not actually visible in the diagram) will be to arrange that the link L U R, the union of L := the "triangular" meridians m, and R := the "rectangular" components of Fig. 5.2 is a rather weak link. Recall that in our construction of M2 a large number K was fixed. Using Lemma 2.8, weakly homotope the "rectangular punctures" R of c in ?M, extending c continuously in the normal direction, so as to make the link L U R have vanishing i/-invariants of length < K. This creates new (c,c)-intersections which are removed as before. Choose the number K such that K > (1 + #,))n + 2c
(5.1)
where #a is the number o f group elements in rcLN represented in the double points of 6 and c is the number of components of R. In moving R' to R many new double points o f c (and thus 6) are created and we have no way to estimate the number. However since ~0M C N is 7h-null and we count group elements only in the group TciN, at most 2 . (~) distinct group elements arise fi'om these double points, and precisely these same elements arise when the (c,c)-intersections are transtbrmed into (6,6)-intersections. Thus it is possible to pick the number K early in the construction of M2 (as we did) tbr the necessary value can be estimated from the number of components ( = c) of {" (-I M and the original double points of 6. We can now turn to the construction of a n M3 satisfying ?M3 ~ (')M2 = Pneib(MU6Uc) and M3 ~ V S I 9 The first point to address is that in Fig.5.1 we formally changed 0-framed 2-handles tt, to l-handles p, (note the dot) as in the passage from Fig.4.1 to Fig.4.2. To justify this we must find slice disks for these components. This requires an S 2 x S2-stablization which we will remove later. Consider the visible (band-like) Seifert surfaces T = {TI . . . . . T,,} for ILl . . . . . It,, in Fig. 5.1. These meet ml . . . . . m,, dually (6,j) and would be suitable for cancellation if they were disks. Push the interiors of Ti slightly into the contractible manifold Z. Let (~,[~) denote the obvious symplectic basis; this is actually the one fixed earlier on 6. Because of the 0-framings, we may perform along {c~} an abstract surgery of pairs o n (Z,T pushed) to obtain ( Z # S 2 x S2's,D) where D is a disjoint union o f n imbedded slice disks tbr lq . . . . . I~,. Morally, M2~ is M2 stablized by these surgeries. To be precise, Fig. 5.1 finally has a meaning as a 4-manifold M + since we now have a place, Z + := Z # S 2 • S2's, to locate the slices indicated by the dots on l q , . . . , l ~ , , . We assume without loss of generality that each 6L.... ,6, has at least one selfintersection as shown in Fig. 5.1. This enables us to compute
~,(M[).
554
M.H. Freedman, P. Teichner
Lemma 5.1. ~ l ( M ] ) Ls" ./reely ,qenerated by the l-handles g f Fi~,t. 5.1 (We exclude here the pseudo-l-handles /tl . . . . . ttn which bound slices in Z + aml are not technically l-handles. )
P r o o f For each its, l __< i :< n, choose a 2-handle from a hyperbolic pair generated by resolving a selfintersection of 6,. Cancel this 2-handle with tL,. Literally this means filling in the slice under tt, with the handle. Fig. 5.1 loses the canceled components and the partner of the canceled 2-handle is joined to a parallel copy of It, by a band. In this reconstituted Fig. 5.1 the attaching regions of all 2-handles can be homotoped off" the standard disks spanning the 1-handles. These homotopies, because of the "rectangles" in Fig. 5.1, do not exist in s but rather in Z+\(1-handle slice disks). After these homotopies we see a homotopy equivalent space whose fundamental group is as claimed. L~ The homology of Z + is conveniently represented in the complement of D by singular spheres of types A and B. An A-sphere has "northern" hemisphere the core-D e bounding (a parallel of) e provided by surgery and "southern" hemisphere a null homotopy of ~ descending further into Z. A B-sphere is made from the toms of length 2~: normal vectors of T C Z restricted to [4 by removing an annular neighborhood around the "lowest" longitudinal copy of/J and gluing in two "southern" null homotopies of [~ descending into Z. In order to get the above torus (and thus the B-sphere) inside Z +, the support of the abstract surgery (turning Z into Z +) should be in an ~:-neighborhood of ~. Because p, and mi, 1 <_ i < n, cancel homologically, the second homology of M2+ is freely generated by singular spheres of types A,B, and C where the spheres of type C are constructed as follows: Consider the n singular punctured spheres Vi. . . . . V,, in M2+ made by capping off "southern" null homotopies for ml . . . . . m,, in Z with "northern" core disks of the attached 2-handles. Each V, acquires a single puncture where it crosses the slice disk for it,. To construct a sphere of type C resolve the puncture on a parallel copy of some V/ by tubing along it i to the longitude of the attaching circle of one of the 0-framed 2-handles linking it~ which were introduced (as Hopf links) to resolve the clasps (coming from the double points of 6), see Fig. 5.1. In this way we get two spheres of type C for each such Hopf link. If all the "southern" null homotopies A in Z (more precisely Z\collar (?Z) used in the construction of A,B, and C are disjoint then (5.2)
A tO B U C is a rtl-null collection of spheres representing the basis of a hyperbolic form in fez(M+).
Actually, less disjointness is required to obtain (5.2). Let A AA U A8 U Ac be the null homotopies needed to form "southern" portions for classes A,B, and C respectively. We make Ac from many parallel copies of a collection Ac, o of disjointly immersed disks described below. Each C-hyperbolic pair derives from a double point of 6. The set .9~ of elements of 7z]N represented by these double points has cardinality #~, see (5.1). Assume that Ac, o are disjoint null homotopies in Z for (1 +#,3) parallel copies of L = {ml . . . . . m,,}, divided into =
4-Manifold topology I1: Dwyer's filtration and surgery kernels
555
Ac,0 = AC, 1 0 U AzC, 0 where AC, I 0 consists of null hornotopies on one copy of L and A2,0 consists of null hornotopies on #~ copies of L. Arbitrarily divide each C-hyperbolic pair into a first and second partner. This division produces, on southern hemispheres, Ac = A~ U A2,. Assume all the A~, are made from near parallel copies of AC, 1 0 and also that A2. consists of near parallel copies of c,0. It is important to make precise which copy of A2c,0 these additional null homotopies should be parallel to. The choice is made by looking at the group element in ~jN of the corresponding double point of 3 and using a bijection o f / f with the parallels of L. Finally assume that all null homotopies AA and A~ are disjoint from each other and A~;0 (and thus Ac). With these conditions (5.2) continues to hold, but the required collection of disjoint null homotopies A0 := Ac~oUAA UAB C A, is much smaller. In fact, card(At, o) _-< (I + #~)n. Furthermore, ~Ac, o is made from parallel copies of the link L and (~(AAU AB) is made from at most 2 parallel copies of the link R by "fusing" (banding together) certain pairs of components (two copies of R are needed to produce ~AB). Thus A0 may be constructed if the link made by taking (1 + #,~) parallel copies of L and 2 parallel copies of R is 4D-homotopically trivial in Z. But by Lemma 2.8 this is assured by the condition that L U R has vanishing ~-invariants of length < (1 + #~)n + 2c, compare (5.1). We finish as in Sects. 3 and 4: Use [FQ, Chapter 6] to s-cobord A U B U C to disjointly imbedded hyperbolic pairs and remove these by surgery. The result M~ of surgery is homotopy equivalent to a wedge of circles with generators corresponding to the meridians to all the one-handles in Fig. 5.1 (excluding lq . . . . . It,,). The manifold N ~ = ( M \ M I ) U M 3 is shown to be simply homotopy equivalent to N + by the direct analogs of diagram 4.4, to compute gl, and lines (4.3) and (4.4) to compute homology. [] 6. The normal bordism This section gives the proof of Corollary 1.2. We suppress the boundaries ~?N and c~X from the notation since they play only a small role. Given a degree l normal map f : N --+ X inducing ~l-isomorphism and kernel Kz represented by M, construct N + and N' as in Sect. 5. Consider following diagram v(N)
N
t~"
I,
t
~ X
) P
N'
~h~
h4
"
N +
556
M.H. Freedman, P. Teichner
Above (hi,h2) and (h3,h4) are two pairs o f (simple) homotopy inverses. The bundle ~ has structure group TOP, v(N) is the TOP stable normal bundle to N, and q : = (h4 o h l ) * ~ . We next check that q is isomorphic to the normal bundle of N ' (with the isomorphism covering idx, ). To compare pull v ( N ' ) and ~ back to N+; h~'~ and h~v(N') are identical over N\MI so the difference is measured in Hi(MI, ()MI; ;c, BTOP) ~ H4-t(MI; ~z,BTOP). Since MI has the integral homology o f a wedge o f circles and Ir3BTOP = 0 the only possible difference lies in Ho(MI; rr4BTOP) ~ 7/. But this obstruction is associated to the first Pontrjagin class or (eight times) the signature and must vanish since sign(X) = sign(N/). It follows that hT?, "~ h~v(N r) and therefore q ~ v(N'). v(N) q v(N) t t Certainly g : N --+ N ~ is equivalent to f : N --+ X, so to prove Corollary 1.2 - that f is normally cobordant to a (simple) homotopy equivalence - it suffices to show that g is normally cobordant to id/v,. Since sign(N) = sign(N ~) the nolTnal invariant n o f g lies in H2(N';77/2). Since g and idN, agree over ( N ' \ M 3 ) the image i* (n) = 0 C Ha(N~\M3;77/2). On the other hand, combining the exact sequence o f the pair with Lefschetz Duality and excision we obtain
H2(N ', Nt\M3; 77/2)
~
H2(N'; 77/2)
l* -----+ H2(N'\M3; 77/2)
II H2(M3,(gM3;77/2)
~-~ H2(M3;2~/2) = 0
This shows that i* is an injection. Thus n = Corollary 1.2.
0 E H2(N';77/2)
proving L~
References A. Casson, M.H. Freedman: Atomic surgery problems. Cont. Math. 35, 181 199 (1984) [c] T. Cochran: Derivative's of links: Milnor's Concordance invariants and Massey products, Vol. 84, # 427, Memoirs of the A.M.S., Providence, Rhode Island, 1990 [D] W. Dwyer: Homology, Massey products and maps between groups. J. Pure Appl. Algebra 6, 177-190 (1975) [FI] M.H. Freedman: The topology of 4-dimensional manifolds. J. Diff. Geom. 17, 357-453 (1982) [F2] M.H. Freedman: A new technique for the link slice problem. Invent. Math. 80, 453-465 (1985) M.H. Freedman: Whitehead3 is a slice link. Invent. Math. 94, 175-182 (1988) [F3] [FQ] M.H. Freedman, F. Quinn: The topology of 4-manifolds. Princeton Math. Series 39, Princeton, NJ, 1990 [FT] M.H. Freedman, P. Teichner: 4-Manifold Topology 1: Subexponential Groups (to appear) [K] Kojima: Milnor's fi-invariants, Massey products and Whitney's trick in 4 dimensions. Top. Appl. 16, 43-60 (1983) X.S. Lin: On equivalence relations of links in 3-manifolds. Preprint (1985) [L] [MKS] W. Magnus, A. Karras, D. Solitar: Combinatorial group theory. Dover Publications, New York, 1966 [CF]
4-Manifold topology ll: Dwyer's filtration and surgery kernels [MI] [M2] [St] [T] IV]
557
J. Milnor: Link Groups. Ann. Math. 59, 177 195 (1954) J. Milnor: lsotopy of links, Algebraic geometry and Topology. Princeton Univ. Press, pp. 280 306, 1957 J. Stallings: Homology and central series of groups. J. Algebra 2, 1970 1981 (1965) V.G. Turaev: Milnor's invariants and Massey products, Studies in topology 11. Zap. Naue. Sere Leningrad (LOMI) 66, 189 203, 209 210 (1976) M. Vaughan-Lee: The restricted Burnside problem. London Math. Soc. Monographs New Series 8, Clarendon Press, 1993